[API Feature Request] Expose current Raycast AI models (GPT-5.6, Gemini 3.5, etc.) through AI.Model

Description

The set of models exposed by the Raycast Extension API through AI.Model is significantly behind the models already available in Raycast AI's product model catalogue and model pickers.

This is not limited to Gemini 3.5 Flash. As of July 10, 2026, I verified the mismatch against the current Raycast model catalogue and the latest published package (@raycast/api@1.104.22).

Examples currently missing from AI.Model

The following models are shown in Raycast AI but do not have corresponding documented, non-deprecated AI.Model members in @raycast/api@1.104.22:

Provider	Models shown in Raycast AI but missing from AI.Model
OpenAI	GPT-5.6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.5, GPT-5.5 Instant, GPT-5.1 Reasoning, GPT-5.2 Reasoning, GPT-5.4 Reasoning
Google	Gemini 3.5 Flash
Anthropic	Claude Haiku 4.5 Reasoning, Claude Sonnet 4.6 Reasoning, Claude Sonnet 5, Claude Opus 4.8
Perplexity	Sonar Reasoning Pro
xAI	Grok-4.5, Grok-4.3
Moonshot AI	Kimi K2.6 (Baseten), Kimi K2.7 Code (Baseten)
MiniMax	MiniMax M2.5 (Baseten)
Z.ai	GLM-5 (Baseten), GLM-5.2 (Baseten)
DeepSeek	DeepSeek V4 Pro (Baseten)

For the GPT-5.6 family, OpenAI's documented model IDs are:

• gpt-5.6-sol — flagship capability; the gpt-5.6 alias routes here
• gpt-5.6-terra — balance of intelligence and cost
• gpt-5.6-luna — efficient, high-volume workloads

Current limitation

For example, these model selections cannot currently be expressed using the documented TypeScript API:

import { AI } from "@raycast/api";

await AI.ask("Review this implementation", {
  model: AI.Model["OpenAI_GPT-5.6_Sol"],
});

await AI.ask("Summarize this document", {
  model: AI.Model["Google_Gemini_3.5_Flash"],
});

The exact enum member names are of course up to Raycast; the examples above follow the existing naming convention.

Proposed outcome

Short term

Please expose the models that are already selectable in Raycast AI through official AI.Model members, starting with the GPT-5.6 family and Gemini 3.5 Flash.

Longer term

To avoid repeating this mismatch whenever providers release new models, please consider one of the following:

1. Keep AI.Model generated and synchronized with the Raycast AI model catalogue.
2. Clearly document that the Extension API intentionally exposes a curated subset, including the eligibility criteria and expected rollout timing.
3. Provide a model-discovery API (for example, AI.getModels()) or another documented way for extensions to select currently enabled models without waiting for a package release.

It would also be helpful to document:

• The minimum Raycast and @raycast/api versions required for each model.
• Effective context limits when models are called through Raycast.
• Model/tool compatibility and whether reasoning variants require additional options not currently present in AI.AskOptions.
• If feasible, the model actually used after fallback. The API reference states that unavailable or disabled models fall back to a similar one, which makes model-specific evaluation and debugging difficult.

Who will benefit from this feature?

Extension developers using AI.ask or useAI who need predictable model behavior for evaluation, latency, cost, tool-use, and output-quality requirements.

Calling provider APIs directly is possible, but it requires every extension to add its own API-key preferences and provider integrations, bypassing the main benefit of Raycast's AI API: no separate keys, configuration, or dependencies.

References

• Raycast AI model catalogue: https://www.raycast.com/core-features/ai/models
• Raycast AI API reference: https://developers.raycast.com/api-reference/ai
• npm package: https://www.npmjs.com/package/@raycast/api
• OpenAI GPT-5.6 model guide: https://developers.openai.com/api/docs/guides/latest-model
• Google Gemini 3.5 Flash documentation: https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash

Start by locating the AI.Model, AI.ask, and useAI declarations in @raycast/api@1.104.22, then compare them with the Raycast AI model catalogue and API reference. Determine whether the outcome is synchronized model members, documented curation, or model discovery, and define the supported models, versions, limits, and fallback behavior before considering the work done.